International Research Journal of Engineering and Technology (IRJET)
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net
Helmet Detection using Machine Learning and Automatic License
Plate Recognition
Lokesh Allamki1, Manjunath Panchakshari2, Ashish Sateesha3, K S Pratheek4
1,2,3,4BNMIT,
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Motorcycle accidents have been rapidly growing
through the years in many countries. In India more than 37
million people use two wheelers. Therefore, it is necessary to
develop a system for automatic detection of helmet wearing
for road safety. Therefore, a custom object detection model is
created using a Machine learning based algorithm which can
detect Motorcycle riders. On the detection of a Helmetless
rider, the License Plate is extracted and the Licence Plate
number is recognized using an Optical Character Recognizer.
This Application can be implemented in real-time using a
Webcam or a CCTV as input.
Key Words: Automatic License Plate Recognition (ALPR),
Deep Neural Network (DNN), Helmet Detection, Machine
Learning, Mean Average Precision (mAP), Optical
Character Recognition (OCR), You Only Look Once
(YOLO).
1. INTRODUCTION
The main safety equipment of motorcyclist is the helmet. The
helmet protects the motorcyclist against accidents. Although
the helmet use is mandatory in many countries, there are
motorcyclists that do not use it or use it incorrectly. Over the
past years many works have been carried out in traffic
analysis, including vehicle detection and classification, and
helmet detection. Intelligent traffic systems were
implemented using computer vision algorithms, such as:
background and foreground image detection to segment the
moving objects in scene and image descriptors to extract
features. Computational intelligence algorithms are used too,
like machine learning algorithms to classify the objects.
Machine learning (ML) is the field of Artificial Intelligence in
which a trained model works on its own using the inputs
given during training period. Machine learning algorithms
build a mathematical model of sample data, known as
"training data", in order to make predictions or decisions
and are also used in the applications of object detection.
Therefore, by training with a specific dataset, a Helmet
detection model can be implemented. Using this helmet
detection model helmet-less riders can be easily detected.
Based one the detected classes the license plate of the rider
is cropped out and saved as an image. This image is given to
an Optical Character Recognition (OCR) model which
recognizes the text and gives the License Plate number as
output in the form of Machine encoded text. And it can also
be implemented in real time using a Webcam.
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal
e-ISSN: 2395-0056
p-ISSN: 2395-0072
BENGALURU
The objective of this paper is to develop a system to enforce
helmet wearing with the help of CCTV cameras. The
developed system aims in changing unsafe behaviors and
consequently reducing the number of accidents and its
severity.
2. RELATED WORK
Over the past years, multiple approaches have been
proposed to solve the problem of helmet detection. The
authors in [7] use a background subtraction method to
detect and differentiate between moving vehicles. And they
used Support Vector Machines (SVM) to classify helmets and
human heads without helmets. Silva et al. in [9] proposed a
hybrid descriptor model based on geometric shape and
texture features to detect motorcyclists without helmet
automatically. They used Hough transform with SVM to
detect the head of the motorcyclist. Additionally, they extend
their work in [10] by multi-layer perception model for
classification of various objects.
Wen et al. [10b] uses a circle arc detection method based
upon the Hough transform. They applied it to detect helmet
on the surveillance system. The drawback of this work is that
they only use geometric features to verify if any safety
helmet exists in the set. Geometric features are not enough
to find helmets. In [11b] it proposes a computer vision
system aiming to detect and segment motorcycles partly. A
helmet detection system is used, and the helmet presence
verifies that there is a motorcycle. In order to detect the
helmet presence, the edges are computed on the possible
helmet region. The Canny edge detector [12b] is used.
Waranusat et al. [11] proposed a system to detect moving
objects using a k-NN classifier over the motorcyclist’s head
to classify helmet. These models were proposed based on
statistical information of images and had a limitation to the
level of accuracy that could be achieved.
With the evolution of neural networks and deep learning
models there was further improvement in the accuracy of
classification. Alex et al. [13] introduced a convolutional
neural network (CNN) based method for object classification
and detection. A. Hirota et al. [12] use a CNN for classification
of helmeted and non-helmeted riders. Although they use
CNN, their helmet detection accuracy is poor with limitations
to helmet color and multiple riders on a single motorcyclist.
| Page 80
 International Research Journal of Engineering and Technology (IRJET)
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net
3. PROPOSED METHODOLOGY
For real-time helmet detection, there is a need for accuracy
and speed. Hence a DNN based model You Only Look Once
(YOLO) was chosen. YOLO is a state-of-the-art, real-time
object detection system.
YOLOv3 is extremely fast and accurate and is a huge
improvement over the previous YOLO versions. It also makes
predictions with a single network evaluation unlike systems
like R-CNN which require thousands for a single image. This
makes it extremely fast, more than 1000x faster than R-CNN
and 100x faster than Fast R-CNN [4].
Object detection is the craft of detecting instances of a
certain class, like animals, humans and many more in an
image or video. The Pre-Existing Object Detection API makes
it easy to detect objects by using pretrained object detection
models. But these models detect several Objects which are of
no use to us, therefore in order to detect the necessary
classes a custom object detector becomes necessary.
In order to implement helmet detection and number plate
recognition and extraction, 5 objects need to be detected.
The objects are – Helmet, No Helmet, Motorbike, Person
(sitting on the bike) and License Plate.
There is a need to create a custom object detection model
that is capable of detecting these objects. A collection of
images containing the objects of the classes to be detected
are used as a Dataset. This dataset is then used to train the
custom model. Once the model has been trained, it can be
used to detect these custom objects.
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal
e-ISSN: 2395-0056
p-ISSN: 2395-0072
The training is done by feeding all the captured images with
their annotations. The model extracts the features of each
class from every image with the help of ground truth of the
required classes. For extracting the features and storing
them to recognize those features from other images, we use
a deep learning, classifier based on the convolutional neural
networks. When an image is given to this trained model the
detection of the pretrained class is necessary. A few images
are taken as an example to show the detection capability of
the custom trained model.
3.1 Helmet Detection
The annotated images are given as input to YOLOv3 model to
train for the custom classes. The weights generated after
training are used to load the model. Once this is done, an
image is given as input. The model detects all the five classes
trained. From this we obtain the information regarding
person riding motorbike. If the person is not wearing a
helmet, then we can easily extract the other class
information of the rider. This can be used to extract the
license plate.
3.2 License Plate Extraction
Once the helmetless rider is detected, the associated person
class is detected. This is done by finding whether the co-
ordinates of the no helmet class lie inside the person class or
not. Similarly, the same steps are followed to detect the
associated motorbike and license plate. Once the coordinates
of the License plate are found, it is cropped and saved as a
new image.
| Page 81
 International Research Journal of Engineering and Technology (IRJET)
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net
3.3 License Plate Recognition
The extracted license plate is given to an Optical Character
Recognition (OCR) model. The OCR recognizes text in the
given image and outputs the recognized strings in the
machine-encoded text. The OCR module within will output a
list of predicted license plate numbers along with a
confidence value. The confidence value indicates how
confident it is in recognizing the given license plate
© 2019, IRJET | Impact Factor value: 7.34
e-ISSN: 2395-0056
p-ISSN: 2395-0072
accurately. Then, the license plate recognized with highest
confidence value is stored in a text file for further use.
4. REAL TIME IMPLEMENTATION
4.1 Using Webcam
The webcam can be used as the input device to receive the
image frames for object detection in real-time. Since we are
using YOLOv3-tiny model, it supports up to 220 fps
processing speed.
4.2 Using IP Webcam for Mobile
Mobile camera can be used as the input rather than using the
webcam. This can open up a lot of possibilities as mobile can
be carried and can cover different angles. Doing all this in
real-time is an added advantage. So, from this not only CCTV
footages but a handheld device can be used for obtaining the
footage. Also, the footage from mobile being up-close can
provide a clearer and more readable number plate for the
OCR to give out an accurate number.
5. RESULTS
The model was trained on tiny YOLOv3 for 11,000 images on
5 classes for 50,000 iterations. The detections of all the
objects classes was obtained with high precision value and
the mean average precision (mAP) reached a constant max
value of 75% hence the training was stopped at 50,000
iterations.
| ISO 9001:2008 Certified Journal | Page 82
 International Research Journal of Engineering and Technology (IRJET)
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net
A few examples of the input image and the output object
detector are shown in the figure below
The code extracts the License plate from the Object detector
output. The License plate extraction code extracts only from
the motor bikes which has a rider who is not wearing helmet
and discards the License plate of the motor bikes whose
rider has helmet.
The OCR model is able to detect and recognize the License
plates present in an image with an accuracy up to 85 percent.
An example of the recognized license plate are shown in the
following figure.
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal
e-ISSN: 2395-0056
p-ISSN: 2395-0072
6. CONCLUSION
From the results shown above it is evident that the YOLO
object detection is well suited for real-time processing and
was able to accurately classify and localize all the object
classes. The proposed end-to-end model was developed
successfully and has all the capabilities to be automated and
deployed for monitoring. For extracting the number plates
some techniques are employed by considering different
cases such as multiple riders without helmets and designed
to handle most of the cases. All the libraries and software
used in our project are open source and hence is very flexible
and cost efficient. The project was mainly built to solve the
problem of non-efficient traffic management. Hence at the
end of it we can say that if deployed by any traffic
management departments, it would make their job easier
and more efficient.
REFERENCES
[1] Viola and Jones, “Robust Real-time Object Detection”,
IJCV 2001.
[2] Navneet Dalal and Bill Triggs, “Histogram of oriented
gradients for human detection”.
[3] Ross, Jeff, Trevor and Jitendra “Rich feature Hierarchy for
Accurate object Detection”.
| Page 83
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072

[4] Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun, “Fast
R-CNN” (Submitted on 4 Jun 2015 (v1), last revised 6 Jan
2016 (this version, v3)).

[5] Joseph Redmon, Ali Farhadi, “YOLO9000: Better, Faster,

Stronger”, University of Washington, Allen Institute Of AI.
[6] Joseph Redmon, Ali Farhadi, “YOLOv3: An Incremental
Improvement”, University of Washington, Allen Institute of
AI.

[6] Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian

Szegedy, Scott Reed, Cheng – Yang Fu, Alexander C. Berg,
“SSD: Single Shot MultiBox Detector”.

[7] A. Adam, E. Rivlin, I. Shimshoni, and D. Reinitz, “Robust

real-time unusual event detection using multiple fixed-
location monitors,” IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol. 30, no. 3, pp. 555–560, March
2008.

[8] AlexeyAB,
https://github.com/AlexeyAB/darknet#requirements.

[9] C.-Y. Wen, S.-H. Chiu, J.-J. Liaw, and C.-P. Lu, “The safety
helmet detection for atm’s surveillance system via the
modified hough transform,” in IEEE 37th Annual
International Carnahan Conference on Security Technology.,
2003, pp. 364–369.

[10] C.-C. Chiu, M.-Y. Ku, and H.-T. Chen, “Motorcycle

detection and tracking system with occlusion segmentation,”
in WIAMIS ’07, USA, 2007

[11] A. Hirota, N. H. Tiep, L. Van Khanh, and N. Oka,

Classifying Helmeted and Non-helmeted Motorcyclists. Cham:
Springer International Publishing, 2017, pp. 81–86.

© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 84